U.S. patent application number 15/691646 was filed with the patent office on 2017-12-21 for system including hierarchical memory modules having different types of integrated circuit memory devices.
The applicant listed for this patent is Rambus Inc.. Invention is credited to Craig Hampel, Mark Horowitz.
Application Number | 20170365354 15/691646 |
Document ID | / |
Family ID | 39512709 |
Filed Date | 2017-12-21 |
United States Patent
Application |
20170365354 |
Kind Code |
A1 |
Hampel; Craig ; et
al. |
December 21, 2017 |
SYSTEM INCLUDING HIERARCHICAL MEMORY MODULES HAVING DIFFERENT TYPES
OF INTEGRATED CIRCUIT MEMORY DEVICES
Abstract
Volatile memory devices may be on a first memory module that is
coupled to a memory controller by a first signal path. A
nonvolatile memory device may be on a second memory module that is
coupled to the first memory module by a second signal path. A
memory transaction for the nonvolatile memory device may be
transferred from the memory controller to at least one of the
volatile memory devices using the first signal path and data
associated with the memory transaction is to be written from at
least one of the volatile memory devices to the nonvolatile memory
device using the second signal path and a control signal. A
durability circuit may generate the control signal based on a
comparison of a number of write transactions to a particular memory
location with a threshold value.
Inventors: |
Hampel; Craig; (Los Altos,
CA) ; Horowitz; Mark; (Menlo Park, CA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Rambus Inc. |
Sunnyvale |
CA |
US |
|
|
Family ID: |
39512709 |
Appl. No.: |
15/691646 |
Filed: |
August 30, 2017 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
15282932 |
Sep 30, 2016 |
9767918 |
|
|
15691646 |
|
|
|
|
14883916 |
Oct 15, 2015 |
9460021 |
|
|
15282932 |
|
|
|
|
12529323 |
Aug 31, 2009 |
9195602 |
|
|
PCT/US08/57471 |
Mar 19, 2008 |
|
|
|
14883916 |
|
|
|
|
60909359 |
Mar 30, 2007 |
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G11C 2029/3202 20130101;
G06F 13/1684 20130101; G06F 12/0804 20130101; G06F 2212/3042
20130101; G06F 2212/608 20130101; G06F 2212/2022 20130101; Y02D
10/00 20180101; G06F 12/0851 20130101; G06F 2212/205 20130101; G11C
29/12015 20130101; G06F 3/0611 20130101; G11C 29/32 20130101; G06F
3/0647 20130101; G06F 12/08 20130101; G11C 5/04 20130101; G06F
3/0688 20130101; G11C 29/76 20130101; G11C 7/10 20130101; G11C
29/1201 20130101; G11C 7/1003 20130101; G11C 29/12 20130101; G06F
3/0614 20130101; G06F 12/0897 20130101 |
International
Class: |
G11C 29/12 20060101
G11C029/12; G06F 12/0897 20060101 G06F012/0897; G06F 12/08 20060101
G06F012/08; G06F 12/0804 20060101 G06F012/0804; G06F 13/16 20060101
G06F013/16; G11C 7/10 20060101 G11C007/10; G11C 5/04 20060101
G11C005/04; G11C 29/00 20060101 G11C029/00; G11C 29/32 20060101
G11C029/32; G06F 3/06 20060101 G06F003/06; G06F 12/0846 20060101
G06F012/0846 |
Claims
1. A system comprising: a plurality of volatile memory devices
disposed on a first memory module that is coupled to a memory
controller by a first signal path; a nonvolatile memory device
disposed on a second memory module that is coupled to the first
memory module by a second signal path, wherein a memory transaction
for the nonvolatile memory device is transferred from the memory
controller to at least one of the plurality of volatile memory
devices using the first signal path, and wherein data associated
with the memory transaction is to be written from the at least one
of the plurality of volatile memory devices to the nonvolatile
memory device using the second signal path and a control signal;
and a durability circuit to generate the control signal based on a
comparison of a number of write transactions to a particular memory
location with a threshold value.
2. The system of claim 1, further comprising: a counter circuit to
count the number of write transactions to the particular memory
location.
3. The system of claim 2, further comprising: a register to store
the threshold value; and a comparison circuit to receive the number
of write transactions from the counter circuit and the threshold
value stored at the register and to compare the number of write
transactions with the threshold value.
4. The system of claim 3, further comprising: a mapping circuit to
receive the particular memory location and the control signal and
to assign a new memory location for the data associated with the
memory transaction based on the control signal.
5. The system of claim 1, wherein the control signal indicates to
assign a new memory location for the data associated with the
memory transaction when the number of write transactions to the
particular memory location is at or exceeds the threshold
value.
6. The system of claim 1, wherein the control signal is further
based on a number of write transactions to other memory locations,
the control signal indicating whether the other memory locations
are associated with a fewer number of write transactions than the
number of write transactions to the particular memory location.
7. The system of claim 1, wherein the plurality of volatile memory
devices and the nonvolatile memory device are in a daisy-chained
configuration.
8. A method comprising: receiving a memory transaction for a memory
system comprising a plurality of volatile memory devices disposed
on a first memory module that is coupled to a memory controller by
a first signal path and a nonvolatile memory device disposed on a
second memory module that is coupled to the first memory module by
a second signal path; transferring data associated with the memory
transaction from the memory controller to at least one of the
plurality of volatile memory devices by using the first signal
path; generating a control signal based on a comparison of a number
of write transactions to a particular memory location with a
threshold value; and transferring data associated with the memory
transaction from the at least one of the plurality of volatile
memory devices to the nonvolatile memory device by using the second
signal path and the control signal.
9. The method of claim 8, further comprising: counting the number
of write transactions to the particular memory location.
10. The method of claim 9, further comprising: storing the
threshold value; and comparing the receive the counted number of
write transactions and the stored threshold value and to compare
the number of write transactions with the threshold value.
11. The method of claim 10, further comprising: assigning a new
memory location for the data associated with the memory transaction
based on the control signal.
12. The method of claim 8, wherein the control signal indicates to
assign a new memory location for the data associated with the
memory transaction when the number of write transactions to the
particular memory location is at or exceeds the threshold
value.
13. The method of claim 8, wherein the control signal is further
based on a number of write transactions to other memory locations,
the control signal indicating whether the other memory locations
are associated with a fewer number of write transactions than the
number of write transactions to the particular memory location.
14. The method of claim 8, wherein the plurality of volatile memory
devices and the nonvolatile memory device are in a daisy-chained
configuration.
15. A circuit comprising: a plurality of volatile memory devices
disposed on a first memory module that is coupled to a memory
controller by a first signal path; an interface to a nonvolatile
memory device disposed on a second memory module that is coupled to
the first memory module by a second signal path, wherein a memory
transaction for the nonvolatile memory device is transferred from
the memory controller to at least one of the plurality of volatile
memory devices using the first signal path, and wherein data
associated with the memory transaction is to be written from the at
least one of the plurality of volatile memory devices to the
nonvolatile memory device using the second signal path and a
control signal; and means for generating the control signal based
on a comparison of a number of write transactions to a particular
memory location with a threshold value.
16. The circuit of claim 15, further comprising: a counter circuit
to count the number of write transactions to the particular memory
location.
17. The circuit of claim 16, further comprising: a register to
store the threshold value; and a compare circuit to receive the
number of write transactions from the counter circuit and the
threshold value stored at the register and to compare the number of
write transactions with the threshold value.
18. The circuit of claim 17, further comprising: a mapping circuit
to receive the particular memory location and the control signal
and to assign a new memory location for the data associated with
the memory transaction based on the control signal.
19. The circuit of claim 15, wherein the control signal indicates
to assign a new memory location for the data associated with the
memory transaction when the number of write transactions to the
particular memory location is at or exceeds the threshold
value.
20. The circuit of claim 15, wherein the control signal is further
based on a number of write transactions to other memory locations,
the control signal indicating whether the other memory locations
are associated with a fewer number of write transactions than the
number of write transactions to the particular memory location.
Description
CROSS-REFERENCE TO RELATED APPLICATION
[0001] This application is a continuation of U.S. patent
application Ser. No. 15/282,932, filed on Sep. 30, 2016, which is a
continuation of U.S. patent application Ser. No. 14/883,916, filed
on Oct. 15, 2015, which is a continuation application of and claims
priority to U.S. patent application Ser. No. 12/529,323 filed on
Aug. 31, 2009, which is a national stage entry of International
Application No. PCT/US08/57471 filed on Mar. 19, 2008, which claims
priority to U.S. Provisional Application No. 60/909,359 filed on
Mar. 30, 2007, each of which is incorporated by reference
herein.
FIELD
[0002] The disclosure herein generally relates to integrated
circuit devices and/or high speed signaling of such devices.
BACKGROUND
[0003] There are significant differences in the performance,
durability, density, cost and power consumption of various memory
technologies and memory systems that implement those memory
technologies. While a particular memory technology may have a
relatively short latency or read access time, the same memory
technology may have a relatively longer write time that may not be
suitable for some applications. A particular memory technology may
be limited to a relatively low number of write operations to a
particular memory location. After exceeding the limited amount of
write operations, information may not be reliably stored and
retrieved from the memory location. A memory technology may be
approximately four to ten times as dense, or occupy a much smaller
surface area/volume, than other memory technologies. Some memory
technologies cost approximately half as much as others. A variety
of voltages or current may be used during memory access operations
in different memory technologies that lead to different power
consumption rates. Thus, some memory technologies use more power
than others.
BRIEF DESCRIPTION OF THE DRAWINGS
[0004] Embodiments are illustrated by way of example, and not by
way of limitation. In the figures of the accompanying drawing, like
reference numerals refer to similar elements.
[0005] FIG. 1 illustrates a hierarchical memory system organized
into logical and physical hierarchies based on memory module
placement and device-type.
[0006] FIG. 2A illustrates an integrated circuit buffer device
having a hierarchy circuit.
[0007] FIG. 2B illustrates an integrated circuit buffer device
similar to FIG. 2A.
[0008] FIG. 3A illustrates relative data mapping between modules of
different hierarchies.
[0009] FIG. 3B illustrates relative data mapping between modules of
different hierarchies.
[0010] FIG. 4 illustrates a controller having a hierarchy
circuit.
[0011] FIG. 5 A-D are flow charts that illustrate methods of
operating a memory system having a hierarchy of memory modules.
DETAILED DESCRIPTION OF EMBODIMENTS
[0012] A memory system, among other embodiments, includes a
controller and hierarchies of memory modules having different types
of integrated circuit memory devices. A hierarchy (of memory
modules) includes one or more memory modules having a particular
type of integrated circuit memory device. Hierarchies of memory
modules may increase overall system performance as compared to a
system having memory modules with a single type of integrated
circuit memory device. Overall system read latencies may be reduced
and write data durability may be increased by using a first type of
integrated circuit memory device in a first hierarchy, while
overall cost and power consumption are reduced by using a second
type of integrated circuit memory device that costs less and
consumes less power in a second hierarchy. For example, at least
one volatile memory device is disposed in a first hierarchy and at
least one non-volatile memory device is disposed in the second
hierarchy so that the first hierarchy may act as a read/write cache
for the controller and second hierarchy.
[0013] In embodiments, hierarchies are coupled in a daisy chain
manner. A first signal path couples a controller to a first memory
module having volatile integrated circuit memory devices. A second
signal path couples the first memory module to a second memory
module having non-volatile memory devices. Control information and
write data to be stored in the volatile memory devices are
transferred, by the controller, on the first signal path. Control
information and write data for the non-volatile memory devices are
transferred from the controller onto the first signal path and then
onto the second signal path by way of the first memory module.
Likewise, read data stored on volatile and non-volatile memory
devices are accessed by the controller from the first signal path
and the second signal path by way of the first memory module.
Control information, read data and write data may be transferred
between the first and second signal paths by an integrated circuit
buffer device disposed on the first memory module.
[0014] Methods of operating a memory system having hierarchies of
memory modules include, among other embodiments, buffering/caching
a block of write data that has been stored in a second hierarchy in
a first hierarchy. The block of write data then may be rewritten
(or refreshed/restored) in the second hierarchy by transferring the
cached block of write data in the first hierarchy to the second
hierarchy in response to a control signal. Read data from the
second hierarchy may be transferred in blocks to the first
hierarchy while a controller accesses data stored (either read or
write data) in the first hierarchy. Write data may be remapped to
different memory locations in the hierarchies to distribute and
minimize write durability. Write data to be stored in a second
hierarchy may be stored in a first hierarchy and read by a
controller while a block of write data is transferred from the
first hierarchy and stored in the second hierarchy. Write data to
be stored in the second hierarchy may be remapped and stored in the
first hierarchy or a different memory location in the first
hierarchy when a defective memory location is detected.
[0015] FIG. 1 illustrates one embodiment of a fully buffered memory
system 100 that employs a memory controller 110, a first set of
memory devices in the form of a first memory module 118 and a
second set of memory devices in the form of a second memory module
120. Respective upstream and downstream signal paths in the form of
point-to-point serial links 140a, 140b and 150a, 150b couple the
controller to the memory modules in a daisy-chained configuration.
A clock source 130 distributes a system clock signal to the
controller and memory modules.
[0016] In an embodiment, the memory controller 110 may be similar
to those employed for use with Fully Buffered Dual lnline Memory
Modules (FBDIMM), in that it transmits serialized control, address
and write data signals, and receives serialized read data signals
consistent with the FBDIMM signaling protocol. Generally, this
includes transmission circuitry in the form of serial link
transmitters to drive multiplexed control, address and write data
signals from the controller to the first memory module along the
downstream serial link 140a. Receiver circuitry on the controller
interfaces with the upstream serial link 150a to receive serialized
read data from the first module 118. A hierarchy circuit 180 may be
employed by the controller to manage hierarchical activities as
will be more fully disclosed below.
[0017] With continued reference to FIG. 1, the first memory module
118 includes a buffer device in the form of an integrated circuit
(IC) 125, and a plurality of volatile memory devices in the form of
dynamic random access memory (DRAM) ICs 103a-h. The buffer IC
includes respective pairs of upstream and downstream ports for
communicating with the controller 110 and the second memory module
120. In one embodiment, the buffer IC includes a hierarchy circuit
190 that may supplement or substitute for the controller hierarchy
circuit 180. The buffer IC communicates with the DRAM devices via
respective control/address and data busses (shown for clarity as a
single bus). The DRAM devices may be of the XDR(n) type, or DDR(n)
type, generally typified by very fast read and write times and high
density.
[0018] Further referring to FIG. 1, the second memory module 120 is
similar to the first memory module 118, with a buffer IC 126 and a
plurality of memory devices 104a-h. However, the memory devices for
the second module are of the nonvolatile type, such as flash
memory. Flash devices are characterized by low cost, low power,
high density, and are capable of reading data at rates similar to
DRAM. However, write operations to a flash device typically involve
erase times that often take milliseconds to complete. Moreover,
flash devices have a limited endurance in terms of data writes to
the same storage blocks. To address this, the buffer IC includes
circuitry that optimizes write operations to the nonvolatile
devices, as will be more fully described below.
[0019] In an embodiment, the first and second modules 118 and 120
are interconnected via the respective buffer ICs in a
point-to-point daisy chain fashion via the upstream and downstream
serial links 140 and 150. In this manner, data necessarily flows
through the first module 118 in any transaction involving the
controller 110 and the second module 120. By positioning the set of
volatile memory devices closer to the controller than the set of
nonvolatile devices, respective first and second level hierarchies
may be defined at 101 and 102.
[0020] To support operation of the system in accordance with the
hierarchies defined by the position and device-type of each module,
the buffer ICs 125 and 126 for the volatile and nonvolatile memory
modules include logic to assist the memory controller 110 in
coordinating transactions, as described more fully below.
[0021] FIG. 2A illustrates an IC buffer device 200a having an HRCH
circuit 190 in an embodiment. IC buffer device 200a corresponds to
the buffer device 126 shown in FIG. 1, and employed by one or more
nonvolatile memory modules 120. In alternate embodiments, one or
more circuits of HRCH circuit 190 may be disposed in controller 110
or distributed between controller 110 and IC buffer device 126. IC
buffer device 200a may be disposed on a memory module, housed in a
common package along with IC memory devices or dies, or situated on
a motherboard, for example, main memory in a personal computer or
server. The IC buffer device 200a may also be used in an embedded
memory subsystem, for example such as one found on a computer
graphics card, video game console or a printer.
[0022] With continued reference to FIG. 2A, an interface 210 is
shown that receives control information, write data and read data
from signal path 201 and outputs control information, write data
and read data to signal path 202 to either nonvolatile (associated
with the second level hierarchy) or volatile integrated circuit
memory devices (associated with the first level hierarchy). In an
embodiment, signal path 201 corresponds to signal paths 140a-b
shown in FIG. 1, while signal path 202 corresponds to signal paths
150a-b. In an embodiment, a multiplexed combination of control
information, write data and read data intended for integrated
circuit memory devices coupled to IC buffer device 200a is received
via interface 210, which may, for example extract the control
information. For example, memory commands and address information
may be decoded and separated from multiplexed information on signal
path 210 and provided to request and address circuit 240 from
interface 210. Write data may be provided to interfaces 220a-b by
way of interface 210 and read data from integrated circuit memory
devices may be received at one or more interfaces 220a-b and
provided to interface 210 via multiplexers 230a-b.
[0023] Interfaces 220a-b are shown as separate interfaces but could
be combined. They contain transaction queues 221a-b that are used
to store data that is being sent to or received from the
nonvolatile memory devices. In the case of buffer 200a being
applied to nonvolatile memory devices, the transaction queues store
a transaction and direct the data to a single or limited number of
memory devices. This data organization allows a transaction to be
serviced by writing only a few memory devices as opposed to all of
the devices, thus decreasing the number of write operations being
performed by a transaction. Data being sent to and from the memory
devices is forwarded to a transaction queue 223a in interface 210
before being sent to the upstream and downstream serial interfaces.
Similarly transactions that are received serially from signal path
201, are stored in the transaction queue 223a and then forwarded to
transaction queues 221a-b in interfaces 220a-b before being sent to
the memory devices. In this organization a single transaction on
the serial interface 210 connected to the upstream and downstream
links gets mapped to only one or two of the nonvolatile memory
devices connected to signal path 121 and 122.
[0024] In an embodiment, a clock signal and other information, are
received on signal path 201 or by other signal paths, such as a
serial bus. In an embodiment, interface 210 includes a transmit
circuit or transmitters and a receiver circuit or receivers (or in
combination referred to as transceivers) to output signals on
signal path 202 and receive signals on signal path 201. Similarly,
interfaces 220a and 220b receive and transmit control information,
read data and write data to and from integrated circuit memory
devices via signal paths 121 and 122, respectively. In an
embodiment, interfaces 220a-b include transmitters and receivers to
output and receive signals on signal paths 121 and 122. In
embodiments, transmitters and receivers in interfaces 210 and
220a-b, singly or in combination, are dedicated to or shared with
particular signal lines in signal paths 121 and 122.
[0025] In an embodiment, transmitters and receivers in interfaces
220a-b transmit and receive signals having a standard volatile
memory device signaling characteristic (or protocol), such as a
Synch Nand or Nor FLASH write/read/erase/control signals.
[0026] In an embodiment, interfaces 220a-b include transmitters to
transfer control information on a unidirectional signal path, while
transmitters and receivers for write and read data transfer the
write and read data on a bidirectional signal path. In an
embodiment, transmitters in interfaces 220a-b and 210 singly or in
combination, are output driver circuits to output respective
signals onto signal paths 121, 122 and 401. The output driver
circuit may be pull-up, pull-down and/or push-pull type output
driver circuits.
[0027] According to an embodiment, multiplexers 230a and 230b
perform bandwidth-concentrating operations, between interface 210
and interfaces 220a and 220b, as well as route data from an
appropriate source (i.e., target a subset of signal paths, internal
data cache-cache circuit 292) to an appropriate destination. In an
embodiment, bandwidth concentration involves combining the
(smaller) bandwidth of each signal path in a multiple signal path
embodiment to match the (higher) overall bandwidth utilized in a
smaller group of signal paths. Bandwidth concentration typically
utilizes multiplexing and demultiplexing of throughput between the
multiple signal paths and smaller group of signal paths. In an
embodiment, IC buffer device 200a utilizes the combined bandwidth
of interfaces 220a and 220b to match the bandwidth of interface
210.
[0028] Further referring to FIG. 2A, in one embodiment the HRCH
circuit 190 includes control circuit 290, mapping circuit 291,
cache (or storage) circuit 292, durability circuit 293 (including
one or more stored write threshold values 293a) and defect circuit
294 (including one or more stored defect memory addresses), singly
or in combination. Circuits shown in IC buffer 200a may be coupled
as shown in FIG. 4. Also, one or more circuit components of HRCH
circuit 190 may be disposed in controller 110, in particular HRCH
circuit 180, rather than in HRCH circuit 190 in various
embodiments.
[0029] Control circuit 290 is responsible for providing and
receiving control signals from and to circuits in HRCH circuit 190.
In an embodiment, control circuit 290 may receive control signals
and/or values from an interface 210 via signal path 201. Control
information may be provided in the form of a command from a
controller 110, programmable circuit (such as an SPD device) and/or
another memory module. In an embodiment, control circuit 290 is a
processor, controller unit and/or control logic. Control circuit
290 is coupled to other circuits in HRCH by way of a signal path
similarly to signal path 405 shown in FIG. 4. Read data, write
data, and addresses as well as control signals may be transferred
between circuits using the signal path. In an embodiment, control
circuit 290 also operates IC buffer device 200a and may include a
compression/decompression engine.
[0030] Mapping circuit 291 is responsible for remapping or
reassigning addresses or memory locations associated with
write/read data. Mapping circuit 291 in response to control signals
from control circuit 290 (and/or durability and defect circuits 293
and 294) remaps addresses and data as illustrated in FIGS. 3A-B. In
an embodiment, mapping circuit 291 may include an address
translator circuit. Mapping circuit 291 may remap addresses that
are stored along with associated read/write data in cache circuit
292. Mapping circuit 291 directs data from interface 210 to the
appropriate transaction queues in interfaces 220 and translates
address received from interface 210 into the appropriate device
select and memory address to be sent to interface 220a-b. Generally
in a volatile memory system a single transaction from interface 210
is directed to one or a few memory devices to reduce the number of
write operations required and consecutive address on interface 210
are directed to consecutive locations in a single memory
device.
[0031] Cache circuit 292 is used to store read/write data from a
controller or another memory module. In an embodiment, the
read/write data is stored along with associated addresses to memory
locations in system 100. Read/write data may be stored and
assembled into groups of contiguous words or bytes, such as 64,
128, 256 and/or 512 Kbyte blocks. These blocks then may be
transferred and stored as a group. For example, a 256 Kbyte block
of write data may be stored in cache circuit 292 of buffer device
125b (which was received from controller 110 via signal paths 140a
and 140b) and transferred to memory module 120c via signal path
140c where the write data is stored (or in an embodiment restored
after a previous write). In another embodiment, integrated circuit
memory devices associated with a particular memory module are used
to store blocks of data instead of cache circuit 292.
[0032] Durability circuit 293 is responsible for determining
whether write/read data should be remapped based on the number of
write operations that have occurred at a particular memory location
(or plurality of memory locations) in system 100. In an embodiment,
durability circuit 293 includes a storage circuit, such as one or
more registers that store a write threshold value or limit the
number of writes to a particular memory location before further
writes or storage at that location may not be reliable. In an
embodiment, durability circuit 293 also includes a counter to count
the number of writes to particular memory locations and a compare
circuit to compare a count value with a stored write threshold
value. The compare circuit, in response to the comparison, then may
output a control signal to the control circuit 290 indicating that
a particular memory location should not be written to. Mapping
circuit 291 may then assign a new address to write/read data to be
stored in one or more memory modules. Durability circuit 293 may
also generate control signals indicating that certain memory
locations have not been written to as much as others so mapping
circuit 291 enables distribution of stored write/read data in
non-volatile integrated circuit memory devices.
[0033] Defect circuit 294 is responsible for determining whether
one or more memory locations are defective or are not accurately
storing and outputting information. In an embodiment, defect
circuit 294 stores a plurality of test values 294a that are written
to one or more memory locations in another memory module in
response to a control signal from control circuit 290. The stored
test values in the other memory module then may be read back to IC
buffer device 200a and compared to the stored plurality of test
values 294a by a compare circuit. The compare circuit outputs a
signal that indicates a memory location is defective in response to
the comparison. A control signal then may be output to the mapping
circuit 291 so that further remapping of write/read data will not
include the identified defective memory location. In an embodiment,
defect circuit 294 stores the addresses of the defective memory
locations that may be accessed by control circuit 290 and/or
mapping circuit 291. In an embodiment, the plurality of test values
294a are generated by a pseudo-random number generator, provided by
controller 110 as a test pattern, or are obtained by reading the
contents of another memory module.
[0034] In an alternate embodiment, defect circuit 294 includes a
sense circuit to sense and compare a current/voltage from a memory
location with a stored value representing a current/voltage in
order to determine whether a memory location is defective.
[0035] In various embodiments, IC buffer device 200a including HRCH
circuit 190 operates in different modes of operation. In a first
mode of operation, IC buffer device 200a stores read/write data to
be accessed by another memory module and/or a controller. In a
second mode of operation, IC buffer device 200a assigns an address
for the read/write data in another memory module and transfers the
data and address to the another memory module which then stores the
read/write data.
[0036] Clock circuit 270 includes one or more clock alignment
circuits for phase or delay adjusting internal clock signals with
respect to an external clock. Clock alignment circuits may utilize
an external clock from an existing clock generator, such as clock
source 130, or an internal clock generator to provide an internal
clock, to generate internal synchronizing clock signals having a
predetermined temporal relationship. In an embodiment, clock
circuit 270 includes a phase lock loop circuit or a delay lock loop
circuit. In an embodiment, clock alignment circuits provide an
internal clock signal having a temporal relationship with
transferred or received control information, read data and/or write
data.
[0037] In embodiments, transmitters in interfaces 220a-b (as well
as interface 210) transmit a differential signal that includes
encoded clock information and receivers receive a differential
signal that includes encoded clock information. In an embodiment,
clock circuit 270 extracts the clock information encoded with the
data received by the receiver. Furthermore, clock information is
encoded with data transmitted by the transmitter. For example,
clock information may be encoded onto a data signal, by ensuring
that a minimum number of signal transitions occur in a given number
of data bits.
[0038] Serial interface 274 is an interface to receive/transmit
serial information from or to a controller 110, another IC buffer
device or other configuration circuit in embodiments. The serial
information may include initialization values/signals for IC buffer
device 200a or a memory module. In an embodiment, serial interface
274 is used by a controller 110 to store and/or read values in HRCH
circuit 190, such as one or more write threshold values 293a and/or
test values 294a.
[0039] FIG. 28 illustrates a buffer IC 200b similar to the buffer
IC 200a shown in FIG. 2A, but for use with volatile memory modules
118 (FIG. 1). Most of the circuitry in the two buffer ICs is
similar, and for reasons of brevity will not be re-described.
However, certain modifications to the circuitry may be employed,
such as the omission of the durability and defect circuits 293a and
294a (FIG. 2A). Additionally, the transaction queues and the
mapping circuit handle data mapping differently than that employed
by the nonvolatile module buffer IC 200a.
[0040] Mapping circuit 291 is generally configured to direct data
to and from interface 220 to all of the devices in parallel and a
similar address is sent to all or most of the devices at the same
time, many deices are selected by a single transaction.
[0041] Interfaces 220a-b are shown as separate interfaces but could
be combined. They contain transaction queues 222a-b that are used
to store data that is being sent to or received from the volatile
memory devices. In the case of buffer 200b being applied to
volatile memory devices like a DRAM, the transaction queues store a
transaction that is directed to and distributed to numerous
devices. In this case the combined bandwidth of the devices are
accessed in parallel to maximize the bandwidth of the transaction.
Data being sent to and from the memory devices is forwarded to a
transaction queue 223b in interface 210 before being sent to the
upstream and downstream serial interfaces. Similarly transactions
that are received serially from signal path 201, are stored in the
transaction queue 223b and then forwarded to transaction queues in
222a-b before being sent to the memory devices. In this
organization a single transaction on the upstream or downstream
links in interface 210 accesses multiple memory devices when it is
read or written from memory.
[0042] As noted above, flash memory devices and DRAM memory devices
have many different characteristics. A further difference involves
how data write operations are carried out in the device.
[0043] FIG. 3A illustrates an example of the relative data mapping
between a DRAM-based FBDIMM module 318 versus a flash-based FBDIMM
module 320. In the DRAM-based FBDIMM memory module, DRAM devices
303a-h on the module generally define the data bus width. The
width, in turn, denotes the number of bits capable of being written
to all the DRAMs simultaneously at each clock edge. Each
simultaneous write is carried out at the same address in each DRAM,
as shown figuratively by the shaded areas in the volatile devices
303a-h of FIG. 3A. Consequently, for each clock cycle, a portion of
the data is written to each DRAM. The buffer IC 325 provides
deserialization circuitry (not shown) to convert the serial stream
received from the upstream serial link 340a or downstream serial
link 350b into parallel data streams for transmission along
parallel data paths 328a-h to the individual DRAM devices.
[0044] While the simultaneous parallel write scheme works well in a
DRAM-based memory module, a variety of reasons render such a method
for flash-based modules impractical. Flash devices often erase and
reprogram in units of memory called blocks. Moreover, in order to
store, or write data to a flash device, the block must be empty, or
erased. Consequently in most cases, an erase operation precedes a
write operation. Further, as noted above, flash devices usually
have a limited write endurance.
[0045] Further referring to FIG. 3A, the flash-based FBDIMM module
320 includes a buffer IC 326 that remaps serialized data received
from the downstream serial link 340b in a way that stores the data
within one or more designated blocks 360 (shown as a shaded block),
utilizing a minimum number of devices. This ensures that the write
operations maximize the available storage utilization per block,
and minimizes the number of write operations affecting the
endurance of all the devices. This mapping may also work in the
reverse direction along the upstream path between the two modules.
In such a situation, data from a given block would be remapped to
the multiple DRAMs similar to that shown.
[0046] FIG. 3B illustrates address mapping 362 between a first
hierarchy address space 363 and a second hierarchy address space
364. In an embodiment, the first address space 363 corresponds to
an address space (or amount of addressable memory locations) in
hierarchy 101 (having volatile integrated circuit memory devices)
and hierarchy address space 364 corresponds to the address space in
hierarchy 102 (having nonvolatile ICs). Hierarchy address space 363
includes at least addressable memory locations 366a-n and hierarchy
address space 364 includes at least addressable memory locations
368a-t. In an embodiment, hierarchy address space 364 is
significantly larger than hierarchy address space 363. As described
herein, circuits and/or executable instructions map or assign data
to either hierarchy address space 363 or hierarchy address space
364. In an embodiment, associated data may be read and/or write
data that was at one time assigned an address in one of the
hierarchy address spaces 363 or 364 and was then remapped or
assigned a different address at a different memory location in the
previously assigned hierarchy address space or to a different
memory location in a different hierarchy address space. In an
embodiment, a memory location illustrated in FIG. 3B corresponds to
a block or group of contiguous memory locations.
[0047] For example, data stored or to be stored at memory location
364a is mapped or assigned an address to memory location 363b and
stored at memory location 363b. Similarly, data stored or to be
stored at memory location 363n is mapped an address to memory
location 364t and stored at memory location 364t.
[0048] Data having an assigned memory location 364e may also be
reassigned a memory location 364s and stored at memory location
364s in hierarchy address space 364. The reassigning and storing of
data in hierarchy address space 364 may occur when a predetermined
number of write operations have occurred at a particular memory
location and/or when a particular memory location is determined to
be defective or unreliable.
[0049] FIG. 4 illustrates a controller 110, as shown in FIG. 1,
with HRCH circuit 180. HRCH circuit 180 operates similar to HRCH
circuits 190 shown in FIG. 2A and FIG. 2B. Controller 110 also
includes an interface 401, similar to interface 310, for
transferring control information and read/write data on signal path
140a. Controller 110 also includes a clock circuit 403, also
similar to clock circuit 370, used in synchronization and
transferring control and read/write data. Serial interface 402 is
used along with serial interface 374 in IC buffer devices 125a-b
and 126a-b to provide serial information between controller 110 and
IC buffer devices 125a-b and 126a-b.
[0050] In an embodiment, control circuit 390 of controller 110 or
IC buffer device 300 reads information from a storage circuit, such
as a SPD device, that indicates the number and types of hierarchies
in system 100. Control circuit 390 then may generate the
appropriate control signals in response to the information read
from a SPD device.
[0051] In operation, the system described above may establish
hierarchies based on module position (e.g. upstream versus
downstream) and device type (e.g. volatile versus nonvolatile).
When coupled with an FBDIMM-like infrastructure, including upstream
and downstream paths, such an architecture takes advantage of
volatile and nonvolatile memory benefits, while avoiding issues of
non-compliance.
[0052] FIGS. 5A-D illustrate methods 500, 510, 520, and 530 for
operating system 100 shown in FIG. 1. By taking advantage of the
daisy-chained point-to-point link architecture described above,
many of the methods described below involve multiple operations
carried out concurrently. Moreover, in embodiments, logic blocks
illustrated in FIGS. 5A-D represent the operation of hardware
(e.g., circuits), software (executable instructions), or a user,
singly or in combination. For example, HRCH circuits shown in FIG.
1, singly or in combination, with other circuits may perform the
operations. Other logic blocks that are not shown may be included
in various embodiments. Similarly, logic blocks that are shown may
be excluded in various embodiments. Also, while methods 500, 510,
520, and 530 are described in sequential logic blocks, operations
represented by logic blocks of the methods are completed very
quickly or almost instantaneously and may be performed
concurrently.
[0053] Method 500 illustrated in FIG. 5A involves write data
buffering and begins by transferring write data from the memory
controller 110 into the first hierarchy 101 of one or more memory
modules, at 501. The write data is also accumulated, and may be
assembled into a predetermined number of contiguous bit values or a
block of write data, in the first hierarchy, at 502. The
accumulated write data in the first hierarchy then may be
transferred from the first hierarchy of modules to the second
hierarchy 102 of one or more memory modules, at 503. Write data
stored in the first hierarchy may then be retired, at 504. In
method 500, the first hierarchy acts as a write cache for the
second hierarchy. This is important since the write latency
associated with the nonvolatile memory devices in the second
hierarchy modules may be on the order of milliseconds. By caching
the write data in the first hierarchy while the data transfers from
the first to the second hierarchy, the controller may concurrently
perform other operations without the need to wait for the second
hierarchy module(s).
[0054] FIG. 5B illustrates a method of read data caching 510
wherein the first hierarchy 101 acts as a read cache for read data
(with respect to the controller 110) for the second hierarchy 102.
Method 510 begins by assembling read data into a predetermined
number of contiguous bit values or a block of read data, then
transferring the read data from the second hierarchy, to the first
hierarchy, at 511 and 512. The first hierarchy is concurrently able
to perform write and read operations concurrently with respect to
the controller while the block of read data is being transferred
from the second hierarchy to the first hierarchy, at 513. Another
beneficial method provided by the hierarchical system described
herein involves write data merging. In some instances, a portion of
some desired data may be stored in a module within the first
hierarchy 101, while another portion is already stored in a module
within the second hierarchy 102. Any writes to a block in the
second hierarchy (with nonvolatile devices) involve first erasing
the contents of the entire block. Consequently, in order to avoid
losing the original contents of the block, a form of
"read-modify-write" series of steps may be employed as shown in
FIG. 5C.
[0055] Further referring to FIG. 5C, a method of write data merging
520 involves first reading desired data "DATA A" from a volatile
memory device in the first hierarchy 101 of modules, at 521. The
data may be temporarily stored in one of the volatile memory module
buffer ICs 125 or nonvolatile memory module buffer ICs 126. A block
of desired data "BLOCK B" from a nonvolatile device is then read
from a memory module in the second hierarchy 102, as shown in step
522, and temporarily stored in the same buffer IC that stores the
data DATA A. The data DATA A and BLOCK Bare then merged together by
the buffer circuitry, at step 523. The merged data "DATA A" and
"BLOCK B" is then written back to the second hierarchy into a
single block of data "BLOCK C", at 524.
[0056] FIG. 5D illustrates a method 530 that begins at 531, by
transferring write data A to the first hierarchy 101, where the
write data A is stored or cached into a block of write data before
the block of write data is written to the second hierarchy 102. At
532, the cached block of write data A in the first hierarchy is
accessed. The cached block of write data A is then transferred, at
533, from the first hierarchy to the second hierarchy and stored in
the second hierarchy while data B in the first hierarchy is
concurrently accessed, such as through a read operation. While FIG.
5D illustrates one embodiment of a method involving concurrent
operations, various combinations of concurrent operations are
possible due to the daisy-chained upstream and downstream
point-to-point links 140a-n and 150a-n that interconnect the
controller and buffer devices 110, 125 and 126. For example, read
and/or write operations may be carried out between the controller
110 and the first hierarchy of memory devices concurrently with
write and/or read operations carried out between the first and
second level memory hierarchies.
[0057] Returning to the embodiments illustrated in FIG. 1, memory
modules 120a-d include a substrate with interfaces 123a-d having a
plurality of conducting contacts, such as pins and/or balls, for
coupling to signal paths 140a-d. In an embodiment, a plurality of
contacts, solder balls or pins are included in an interface to
provide electrical connections between an interface and a memory
module substrate. In an embodiment, the interface may be removable
from a connector or substrate, such as a printed circuit board. In
an embodiment, controller 110, integrated circuit memory devices
103a-p and 104a-p, and IC buffer devices 125a-b and 126a-b are
separately housed integrated monolithic circuits and/or combined in
multiple package (e.g., memory devices and a buffer device in a
single package). The packages may be disposed on one or more
substrates.
[0058] In an embodiment, one or more signal paths 140a-d are
point-to-point links that denotes one or a plurality of signal
lines, each signal line having only two transceiver connection
points, each transceiver connection point coupled to a transmitter,
a receiver or transceiver circuit. For example, a point-to-point
link may include a transmitter coupled at or near one end and a
receiver coupled at or near the other end.
[0059] In embodiments, one or more signal paths 140a-d may include
different types of bus or point-to-point link architectures. In
embodiments, signal paths may also have different types of
signaling and clocking type architectures. Embodiments having
different link architectures include simultaneous bi-directional
links, time-multiplexed bi-directional links and multiple
unidirectional links. Voltage or current mode signaling may be
employed in any of these link or bus architectures. In an
embodiment, signal paths 140a-d include a serial data bus, such as
an SMBus. A serial bus may couple controller 110 to one or more IC
buffer devices 125a-b and 126a-b via serial interface 474.
[0060] In an embodiment, one or more memory modules 120a-d are
dual-in-line memory modules ("DIMM") having a standard DIMM form
factor. In an embodiment, a memory module may be included in a
single unitary package, as in a "system in package" ("SIP"). In one
type of SIP embodiment, a memory module may include a series of
integrated circuit dies (i.e., memory devices and buffer devices)
stacked on top of one another and coupled via a conductive
interconnect. Solder balls or wire leads may be employed as the
connector interface such that the memory module may be fixedly
attached to a printed circuit board substrate. A connector
interface may also be of a physically separable type that includes,
for example, male and female portions such that a memory module is
detachable from the rest of a system. Another SIP embodiment may
include a number of memory devices and a buffer device disposed, in
a two dimensional arrangement, on a common substrate plane and
situated inside a single package housing.
[0061] In embodiments shown in FIG. 1, integrated circuit memory
devices 103a-p and 104a-p include different types of integrated
circuit memory devices. For example, integrated circuit memory
devices 103a-p may be volatile integrated circuit memory devices
while integrated circuit memory devices 104a-p may be non-volatile
integrated circuit memory devices. A volatile integrated circuit
memory device includes a plurality of memory cells where stored
information is lost when power is removed from the device in an
embodiment. In contrast, a non-volatile integrated circuit memory
device includes a plurality of memory cells where information is
maintained when power is removed from the device in an
embodiment.
[0062] Types of volatile memory devices include, but are not
limited to, dynamic random access memory ("DRAM"), molecular
charge-based (ZettaCore) DRAM, floating-body DRAM and static random
access memory ("SRAM"). Particular types of DRAM include double
data rate SDRAM ("DDR"), or later generation SDRAM (e.g., "DDR2" or
"DDR3") as well as XDR.TM. DRAM or DIRECT RAMBUS.RTM. DRAM
("DRDRAM").
[0063] Types of non-volatile memory devices include, but are not
limited to, types of electrically erasable program read-only memory
("EEPROM"), FLASH (including NAND and NOR FLASH), ONO FLASH,
magnetoresistive or magnetic RAM ("MRAM"), ferroelectric RAM
("FRAM"), holographic media, Ovonic/phase change, Nanocrystals,
Nanotube RAM (NRAM-Nantero), MEMS scanning probe systems, MEMS
cantilever switch, polymer, molecular, nano-floating gate and
single electron.
[0064] One characteristic of some types of non-volatile memory
devices is that although it can be read or programmed/written a
byte or word at a time in a random access manner, it must be erased
a block at a time. Starting with a freshly erased block, any byte
within the block can be programmed. However, once a byte has been
programmed, it cannot typically be changed again until the entire
block is erased. For example, NOR FLASH memory devices may offer
random-access read and programming operations, but typically cannot
offer random-access rewrite or erase operations.
[0065] An integrated circuit memory device may include one or more
storage arrays (or banks) of two dimensional memory cells that may
be individually accessed (via word and bit lines) by a row and
decoder circuit along with a plurality of sense amplifiers. Various
amounts of voltages or currents may be applied to one or more
memory cells in a storage array to program (write), read and/or
erase a memory cell storing a voltage value corresponding to a
logical value (e.g., logical 1 or 0).
[0066] In embodiments shown in FIG. 1, controller 110 is a master
device, which may be an integrated circuit device that contains
other interfaces or functionality, for example, a Northbridge chip
of a chip set. The master device may be integrated on a
microprocessor or a graphics processor unit ("GPU") or visual
processor unit ('VPU''). The master device may be implemented as a
field programmable gate array ("FPGA"). System 100 may be included
in various systems or subsystems such as servers (e.g., a blade
server), personal computers, graphics cards, set-top boxes, cable
modems, cell phones, game consoles, digital television sets (for
example, high definition television ("HDTV")), fax machines, cable
modems, digital versatile disc ("DVD") players or network
routers.
[0067] In embodiments shown in FIG. 1, system 100 may be used in a
server used in executing or functioning as a search engine that
provides search results including hyperlinks to web sites (or
pages), images and/or videos stored on the Internet in response to
entering key words. In a search engine embodiment, information that
is often retrieved may be stored in hierarchy 101 (e.g., recent
search results) while information that is less often retrieved
(e.g., indexes) may be stored in hierarchy 102.
[0068] Signals described herein may be transmitted or received
between and within controllers/memory modules/devices/circuits by
electrical conductors and generated using any number of signaling
techniques including without limitation, modulating the voltage or
current level of an electrical signal. The signals may represent
any type of control and timing information (e.g., commands, address
values, clock signals, and configuration information) as well as
data. In an embodiment, control, address and or data information is
included in a request packet that represents one or more signals
asserted at particular bit windows on particular signal lines. In
an embodiment, a request packet is provided from a controller to
one or more memory modules and between memory modules.
[0069] In embodiments, a signal path is media that transfers a
signal, such as an interconnect, conducting element, contact, pin,
region in a semiconductor substrate, wire, metal trace/signal line,
or photoelectric conductor, singly or in combination. In an
embodiment, multiple signal paths may replace a single signal path
illustrated in the figures and a single signal path may replace
multiple signal paths illustrated in the figures. In embodiments, a
signal path may include a bus and/or point-to-point connection. In
an embodiment, a signal path includes control and data signal
lines. In an alternate embodiment, a signal path includes only data
signal lines or only control signal lines. In still other
embodiments, signal paths are unidirectional (signals that travel
in one direction) or bidirectional (signals that travel in two
directions) or combinations of both unidirectional signal lines and
bidirectional signal lines.
[0070] It should be noted that the various circuits disclosed
herein may be described using computer aided design tools and
expressed (or represented) as data and/or instructions embodied in
various computer-readable media, in terms of their behavior,
register transfer, logic component, transistor, layout geometries,
and/or other characteristics. Formats of files and other objects in
which such circuit expressions may be implemented include, but are
not limited to: formats supporting behavioral languages such as C,
Verilog, and HLDL; formats supporting register level description
languages like RTL; formats supporting geometry description
languages such as GDSII, GDSIII, GDSIV, CIF, MEBES; and any other
suitable formats and languages. Computer-readable media in which
such formatted data and/or instructions may be embodied include,
but are not limited to, non-volatile storage media in various forms
(e.g., optical, magnetic or semiconductor storage media) and
carrier waves that may be used to transfer such formatted data
and/or instructions through wireless, optical, or wired signaling
media or any combination thereof. Examples of transfers of such
formatted data and/or instructions by carrier waves include, but
are not limited to, transfers (uploads, downloads, e-mail, etc.)
over the Internet and/or other computer networks via one or more
data transfer protocols (e.g., HTTP, FTP, SMTP, etc.). When
received within a computer system via one or more computer-readable
media, such data and/or instruction-based expressions of the above
described circuits may be processed by a processing entity (e.g.,
one or more processors) within the computer system in conjunction
with execution of one or more other computer programs including,
without limitation, netlist generation programs, place and route
programs and the like, to generate a representation or image of a
physical manifestation of such circuits. Such representation or
image may thereafter be used in device fabrication, for example, by
enabling generation of one or more masks that are used to form
various components of the circuits in a device fabrication
process.
[0071] The functions of circuits described herein may also be
replaced partially or wholly with computer/processor executable
instructions (or software) stored and executed in for example
system 100.
[0072] The foregoing description of the preferred embodiments has
been provided for the purposes of illustration and description. It
is not intended to be exhaustive or to limit the embodiments to the
precise forms disclosed. Modifications and variations will be
apparent to practitioners skilled in the art. The embodiments were
chosen and described in order to best explain the principles of the
invention and its practical applications, thereby enabling others
skilled in the art to understand the invention for various
embodiments and with the various modifications as are suited to the
particular use contemplated. It is intended that the scope of the
invention be defined by the following claims and their
equivalents.
* * * * *